UP | HOME

Control of the NASS with optimal stiffness

Table of Contents

1 Low Authority Control - Decentralized Direct Velocity Feedback

control_architecture_dvf.png

Figure 1: Low Authority Control: Decentralized Direct Velocity Feedback

1.1 Initialization

initializeGround();
initializeGranite();
initializeTy();
initializeRy();
initializeRz();
initializeMicroHexapod();
initializeAxisc();
initializeMirror();

initializeSimscapeConfiguration();
initializeDisturbances('enable', false);
initializeLoggingConfiguration('log', 'none');

initializeController('type', 'hac-dvf');

We set the stiffness of the payload fixation:

Kp = 1e8; % [N/m]

1.2 Identification

K = tf(zeros(6));
Kdvf = tf(zeros(6));

We identify the system for the following payload masses:

Ms = [1, 10, 50];

The nano-hexapod has the following leg’s stiffness and damping.

initializeNanoHexapod('k', 1e5, 'c', 2e2);

1.3 Controller Design

The obtain dynamics from actuators forces \(\tau_i\) to the relative motion of the legs \(d\mathcal{L}_i\) is shown in Figure 2 for the three considered payload masses.

The Root Locus is shown in Figure 3 and wee see that we have unconditional stability.

In order to choose the gain such that we obtain good damping for all the three payload masses, we plot the damping ration of the modes as a function of the gain for all three payload masses in Figure 4.

opt_stiff_dvf_plant.png

Figure 2: Dynamics for the Direct Velocity Feedback active damping for three payload masses

opt_stiff_dvf_root_locus.png

Figure 3: Root Locus for the DVF controll for three payload masses

Damping as function of the gain

opt_stiff_dvf_damping_gain.png

Figure 4: Damping ratio of the poles as a function of the DVF gain

Finally, we use the following controller for the Decentralized Direct Velocity Feedback:

Kdvf = 5e3*s/(1+s/2/pi/1e3)*eye(6);

1.4 Effect of the Low Authority Control on the Primary Plant

Let’s identify the dynamics from actuator forces \(\bm{\tau}\) to displacement as measured by the metrology \(\bm{\mathcal{X}}\): \[ \bm{G}(s) = \frac{\bm{\mathcal{X}}}{\bm{\tau}} \] We do so both when the DVF is applied and when it is not applied.

Then, we compute the transfer function from forces applied by the actuators \(\bm{\mathcal{F}}\) to the measured position error in the frame of the nano-hexapod \(\bm{\epsilon}_{\mathcal{X}_n}\): \[ \bm{G}_\mathcal{X}(s) = \frac{\bm{\epsilon}_{\mathcal{X}_n}}{\bm{\mathcal{F}}} = \bm{G}(s) \bm{J}^{-T} \] The obtained dynamics is shown in Figure 5.

A zero with a positive real part is introduced in the transfer function from \(\mathcal{F}_y\) to \(\mathcal{X}_y\) after Decentralized Direct Velocity Feedback is applied.

And we compute the transfer function from actuator forces \(\bm{\tau}\) to position error of each leg \(\bm{\epsilon}_\mathcal{L}\): \[ \bm{G}_\mathcal{L} = \frac{\bm{\epsilon}_\mathcal{L}}{\bm{\tau}} = \bm{J} \bm{G}(s) \] The obtained dynamics is shown in Figure 6.

opt_stiff_primary_plant_damped_X.png

Figure 5: Primary plant in the task space with (dashed) and without (solid) Direct Velocity Feedback

opt_stiff_primary_plant_damped_L.png

Figure 6: Primary plant in the space of the legs with (dashed) and without (solid) Direct Velocity Feedback

The coupling (off diagonal elements) of \(\bm{G}_\mathcal{X}\) are shown in Figure 7 both when DVF is applied and when it is not.

The coupling does not change a lot with DVF.

The coupling in the space of the legs \(\bm{G}_\mathcal{L}\) are shown in Figure 8.

The magnitude of the coupling between \(\tau_i\) and \(d\mathcal{L}_j\) (Figure 8) around the resonance of the nano-hexapod (where the coupling is the highest) is considerably reduced when DVF is applied.

opt_stiff_primary_plant_damped_coupling_X.png

Figure 7: Coupling in the primary plant in the task with (dashed) and without (solid) Direct Velocity Feedback

opt_stiff_primary_plant_damped_coupling_L.png

Figure 8: Coupling in the primary plant in the space of the legs with (dashed) and without (solid) Direct Velocity Feedback

1.5 Effect of the Low Authority Control on the Sensibility to Disturbances

We may now see how Decentralized Direct Velocity Feedback changes the sensibility to disturbances, namely:

  • Ground motion
  • Spindle and Translation stage vibrations
  • Direct forces applied to the sample

To simplify the analysis, we here only consider the vertical direction, thus, we will look at the transfer functions:

  • from vertical ground motion \(D_{w,z}\) to the vertical position error of the sample \(E_z\)
  • from vertical vibration forces of the spindle \(F_{R_z,z}\) to \(E_z\)
  • from vertical vibration forces of the translation stage \(F_{T_y,z}\) to \(E_z\)
  • from vertical direct forces (such as cable forces) \(F_{d,z}\) to \(E_z\)

The norm of these transfer functions are shown in Figure 9.

opt_stiff_sensibility_dist_dvf.png

Figure 9: Norm of the transfer function from vertical disturbances to vertical position error with (dashed) and without (solid) Direct Velocity Feedback applied

Decentralized Direct Velocity Feedback is shown to increase the effect of stages vibrations at high frequency and to reduce the effect of ground motion and direct forces at low frequency.

1.6 Conclusion

2 Primary Control in the leg space

In this section we implement the control architecture shown in Figure 10 consisting of:

  • an inner loop with a decentralized direct velocity feedback control
  • an outer loop where the controller \(\bm{K}_\mathcal{L}\) is designed in the frame of the legs

control_architecture_hac_dvf_pos_L.png

Figure 10: Cascade Control Architecture. The inner loop consist of a decentralized Direct Velocity Feedback. The outer loop consist of position control in the leg’s space

The controller for decentralized direct velocity feedback is the one designed in Section 1.

2.1 Plant in the leg space

We now look at the transfer function matrix from \(\bm{\tau}^\prime\) to \(\bm{\epsilon}_{\mathcal{X}_n}\) for the design of \(\bm{K}_\mathcal{L}\).

The diagonal elements of the transfer function matrix from \(\bm{\tau}^\prime\) to \(\bm{\epsilon}_{\mathcal{X}_n}\) for the three considered masses are shown in Figure 11.

The plant dynamics below \(100\ [Hz]\) is only slightly dependent on the payload mass.

opt_stiff_primary_plant_L.png

Figure 11: Diagonal elements of the transfer function matrix from \(\bm{\tau}^\prime\) to \(\bm{\epsilon}_{\mathcal{X}_n}\) for the three considered masses

2.2 Control in the leg space

We design a diagonal controller with all the same diagonal elements.

The requirements for the controller are:

  • Crossover frequency of around 100Hz
  • Stable for all the considered payload masses
  • Sufficient phase and gain margin
  • Integral action at low frequency

The design controller is as follows:

  • Lead centered around the crossover
  • An integrator below 10Hz
  • A low pass filter at 250Hz

The loop gain is shown in Figure 12.

h = 2.0;
Kl = 2e7 * eye(6) * ...
     1/h*(s/(2*pi*100/h) + 1)/(s/(2*pi*100*h) + 1) * ...
     1/h*(s/(2*pi*200/h) + 1)/(s/(2*pi*200*h) + 1) * ...
     (s/2/pi/10 + 1)/(s/2/pi/10) * ...
     1/(1 + s/2/pi/300);

opt_stiff_primary_loop_gain_L.png

Figure 12: Loop gain for the primary plant

Finally, we include the Jacobian in the control and we ignore the measurement of the vertical rotation as for the real system.

load('mat/stages.mat', 'nano_hexapod');
K = Kl*nano_hexapod.kinematics.J*diag([1, 1, 1, 1, 1, 0]);

2.3 Sensibility to Disturbances and Noise Budget

We identify the transfer function from disturbances to the position error of the sample when the HAC-LAC control is applied.

We compare the norm of these transfer function for the vertical direction when no control is applied and when HAC-LAC control is applied: Figure 13.

opt_stiff_primary_control_L_senbility_dist.png

Figure 13: Sensibility to disturbances when the HAC-LAC control is applied

Then, we load the Power Spectral Density of the perturbations and we look at the obtained PSD of the displacement error in the vertical direction due to the disturbances:

  • Figure 14: Amplitude Spectral Density of the vertical position error due to both the vertical ground motion and the vertical vibrations of the spindle
  • Figure 15: Comparison of the Amplitude Spectral Density of the vertical position error in Open Loop and with the HAC-DVF Control
  • Figure 16: Comparison of the Cumulative Amplitude Spectrum of the vertical position error in Open Loop and with the HAC-DVF Control

opt_stiff_primary_control_L_psd_dist.png

Figure 14: Amplitude Spectral Density of the vertical position error of the sample when the HAC-DVF control is applied due to both the ground motion and spindle vibrations

opt_stiff_primary_control_L_psd_tot.png

Figure 15: Amplitude Spectral Density of the vertical position error of the sample in Open-Loop and when the HAC-DVF control is applied

opt_stiff_primary_control_L_cas_tot.png

Figure 16: Cumulative Amplitude Spectrum of the vertical position error of the sample in Open-Loop and when the HAC-DVF control is applied

2.4 Simulations of Tomography Experiment

Let’s now simulate a tomography experiment. To do so, we include all disturbances except vibrations of the translation stage.

initializeDisturbances();
initializeSimscapeConfiguration('gravity', false);
initializeLoggingConfiguration('log', 'all');

And we run the simulation for all three payload Masses.

2.5 Results

Let’s now see how this controller performs.

First, we compute the Power Spectral Density of the sample’s position error and we compare it with the open loop case in Figure 17.

Similarly, the Cumulative Amplitude Spectrum is shown in Figure 18.

Finally, the time domain position error signals are shown in Figure 19.

opt_stiff_hac_dvf_L_psd_disp_error.png

Figure 17: Amplitude Spectral Density of the position error in Open Loop and with the HAC-LAC controller

opt_stiff_hac_dvf_L_cas_disp_error.png

Figure 18: Cumulative Amplitude Spectrum of the position error in Open Loop and with the HAC-LAC controller

opt_stiff_hac_dvf_L_pos_error.png

Figure 19: Position Error of the sample during a tomography experiment when no control is applied and with the HAC-DVF control architecture

2.6 Actuator Stroke and Forces

opt_stiff_hac_dvf_L_act_force.png

Figure 20: Force applied by the actuator during the simulation

opt_stiff_hac_dvf_L_act_stroke.png

Figure 21: Leg’s stroke during the simulation

2.7 Conclusion

3 Further More complex simulations

3.1 Simulation with Micro-Hexapod Offset

3.1.1 Simulation

The micro-hexapod is inducing a 10mm offset of the sample center of mass with the rotation axis. A tomography experiment is then simulated.

initializeDisturbances();
initializeSimscapeConfiguration('gravity', false);
initializeLoggingConfiguration('log', 'all');

initializeSample('mass', 1, 'freq', 200);
initializeMicroHexapod('AP', [10e-3 0 0]);
initializeReferences('Rz_type', 'rotating', 'Rz_period', 1, ...
                     'Dh_pos', [10e-3; 0; 0; 0; 0; 0]);
load('mat/conf_simulink.mat');
set_param(conf_simulink, 'StopTime', '3');
sim('nass_model');

3.1.2 Results

opt_stiff_hac_dvf_Dh_offset_disp_error.png

opt_stiff_hac_dvf_Dh_offset_F.png

opt_stiff_hac_dvf_Dh_offset_dL.png

3.2 Simultaneous Translation scans and Spindle’s rotation

3.2.1 Simulation

A simulation is now performed with translation scans and spindle rotation at the same time.

The sample has a mass one 1kg, the spindle rotation speed is 60rpm and the translation scans have a period of 4s and a triangular shape.

initializeDisturbances();
initializeSimscapeConfiguration('gravity', false);
initializeLoggingConfiguration('log', 'all');

initializeSample('mass', 1, 'freq', 200);
initializeReferences('Rz_type', 'rotating', 'Rz_period', 1, ...
                     'Dy_type', 'triangular', 'Dy_amplitude', 5e-3, 'Dy_period', 4);

3.2.2 Results

opt_stiff_hac_dvf_Dy_scans_disp_error.png

opt_stiff_hac_dvf_Dy_scans_F.png

opt_stiff_hac_dvf_Dy_scans_dL.png

4 Primary Control in the task space

In this section, the control architecture shown in Figure 28 is applied and consists of:

  • an inner Low Authority Control loop consisting of a decentralized direct velocity control controller
  • an outer loop with the primary controller \(\bm{K}_\mathcal{X}\) designed in the task space

control_architecture_hac_dvf_pos_X.png

Figure 28: HAC-LAC architecture

4.1 Plant in the task space

Let’s look \(\bm{G}_\mathcal{X}(s)\).

4.2 Control in the task space

Kx = tf(zeros(6));

h = 2.5;
Kx(1,1) = 3e7 * ...
          1/h*(s/(2*pi*100/h) + 1)/(s/(2*pi*100*h) + 1) * ...
          (s/2/pi/1 + 1)/(s/2/pi/1);

Kx(2,2) = Kx(1,1);

h = 2.5;
Kx(3,3) = 3e7 * ...
          1/h*(s/(2*pi*100/h) + 1)/(s/(2*pi*100*h) + 1) * ...
          (s/2/pi/1 + 1)/(s/2/pi/1);
h = 1.5;
Kx(4,4) = 5e5 * ...
          1/h*(s/(2*pi*100/h) + 1)/(s/(2*pi*100*h) + 1) * ...
          (s/2/pi/1 + 1)/(s/2/pi/1);

Kx(5,5) = Kx(4,4);

h = 1.5;
Kx(6,6) = 5e4 * ...
          1/h*(s/(2*pi*30/h) + 1)/(s/(2*pi*30*h) + 1) * ...
          (s/2/pi/1 + 1)/(s/2/pi/1);

4.2.1 Stability

for i = 1:length(Ms)
    isstable(feedback(Gm_x{i}*Kx, eye(6), -1))
end

4.3 Simulation

4.4 Conclusion

Author: Dehaeze Thomas

Created: 2020-05-05 mar. 10:34